Deep Learning Based Image Segmentation for Computer Vision

In this paper, we are exploring deep learning based image segmentation methods and evaluating the performance of different deep learning models in image segmentation tasks. U-Net, DeepLabv3+, FCN and Mask R-CNN models are compared experimentally to analyze their accuracy, computational efficiency and applicability in semantic and instance segmentation. The experimental results show that DeepLabv3+ has an obvious advantage in accuracy, while Mask R-CNN excels in instance segmentation. Despite the difference in accuracy, these models still need to be further optimized in terms of inference speed and training time.